Mistral AI

Mistral Large 4 (Le Chonk)

Open weights

Open-weight natively multimodal mixture-of-experts model. Mistral's Oct. 29, 2026 post releases the weights and prices the API at $0.70 per million input tokens, $0.07 cached, and $2.10 output. It describes a 1-trillion-parameter model and does not restate the active-parameter count. The Hugging Face card, read Nov. 5, 2026, states 49 billion active parameters per token and 52 billion including embeddings and output layers. That reconciles the Oct. 6 figure of 52 billion active parameters with the Oct. 21 figure of 49 billion. The repository lists a LICENSE file and consolidated safetensors shards. The license is the Mistral Ecosystem Open-Weight license, described as modified MIT, and model-as-a-service use needs a separate commercial license.

Latest update Nov 5, 2026

01

Current status

Weights and the license were public on the Hugging Face card when read Nov. 5, 2026. The card states 49 billion active parameters per token and 52 billion including embeddings and output layers. The Oct. 29 weights post does not restate that count. API price on that post is $0.70 input, $0.07 cached input, and $2.10 output per million tokens.

02

Specifications4

FrontierCode 2.0
19.7%
Terminal-Bench 4.0
41%
Artificial Analysis Coding Agent Index
48%
Artificial Analysis Cyber Index
51%
  1. Mistral releases open weights for Large 4

    Mistral released open weights for Mistral Large 4 on Oct. 29, 2026, after a public preview earlier in the month, and priced the API at $0.70 per million input tokens, $0.07 cached, and $2.10 output. The company calls it a 1-trillion-parameter multimodal model and says it leads open-weight models on FrontierCode 2.0 at 19.7% and on Terminal-Bench 4.0 at 41%. The license is a modified MIT license that requires a separate commercial license for model-as-a-service use, and the post names Microsoft Foundry and Nvidia NIM as hosted options.

    mistral.ai + Microsoft+ Nvidia

  2. Mistral and Nvidia plan engineering AI benchmarks

    Mistral and Nvidia said on Oct. 21 they will combine Mistral Large 4 and Mistral's physics-modeling pipelines with Nvidia CUDA-X, Warp, and Omniverse libraries for engineering simulation and design. They plan public benchmarks and open datasets for physics models, plus tests of whether agents can complete computer-aided engineering and chip-design work. Neither post gives a date for those benchmarks, and Mistral again said Large 4 weights are planned for the end of October.

    mistral.ai blogs.nvidia.com + Nvidia

  3. Mistral opens Large 4 preview, weights due end of October

    Mistral launched a public preview of Mistral Large 4 (Le Chonk) on 6 October 2026, describing a 1-trillion-parameter natively multimodal mixture-of-experts model with 49 billion active parameters, trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its European datacenters. The company said the preview API is live on Mistral Studio and that open weights will follow by the end of October after red-teaming with cybersecurity partners and state authorities; Axios, citing Mistral, named 27 October and quoted VP of science Pierre Stock saying the lab is not yet at the closed-model frontier. Independent evaluation of the preview API was not completed this run; SOTA-among-open-weights claims remain company-reported.

    mistral.ai www.axios.com www.theverge.com

04

More from Mistral AI7

05

More in AI Labs